Association of Interpreters and Translators (AIT) – Written Evidence (ITS0047)

House of Lords Public Services Committee 

Inquiry: Interpreting and translation services in the courts

At AIT our aim is to achieve and maintain protection of title and the regulation of the profession by helping to raise the skill and passion of Interpreters and Translators. AIT links members with professional training courses, CPD training, networking opportunities and more. AIT is a member of the International Federation of Translators (FIT), FIT Europe – Regional Centre for Europe of the International Federation of Translators (FIT), the European Legal Interpreters and Translators Association (EULITA), Professional Interpreters for Justice (PI4J), Public Service Interpreters and Translators Group (PSIT) and the MOJ Language Services External Stakeholder Forum.

Contributors: Byron Campbell, Elena Shutova, Gintare Daunoraviciute, Helena El Masri, Irina Sanders, Sophie Man-Cheung and Sue Leschen.

Table of Contents                                                                                                                                             Page

Submission

Appendix 1

Appendix 2

Appendix 3 – Response by FIT Europe to Question 6 and 7 of the Inquiry

Submission

To what extent do the current interpreting and translation services provided in courts meet the needs of those involved in proceedings, including defendants, witnesses, prosecutors and legal professionals?

1. There are multiple issues with court interpreting services, such as interpreters not showing up, being double-booked, sent to the wrong court, or booked for the wrong language or dialect[1]. These problems negatively impact all parties involved, including lawyers, judges, and defendants[2]. Without professional interpreters, lawyers are burdened with explaining proceedings to clients in languages they do not speak[3].
2. Additionally, a lack of clear guidelines regarding interpreters' roles leads to confusion. Judges sometimes expect interpreters to interpret private consultations, which creates conflicts of interest and distrust between parties. Interpreters are also often approached inappropriately by jurors or the public, as they are not always seen as integral to court proceedings like other legal professionals.
3. Judges may fail to instruct juries or the public to refrain from interacting with interpreters, which places interpreters in vulnerable positions. Furthermore, interpreters are sometimes not given sufficient breaks, leading to fatigue and errors. Unlike spoken language interpreters, sign language interpreters work in pairs to avoid such issues.

How have interpreting and translation services changed in recent years?  

4. Prior to 2012, court interpreters were directly booked by the courts under the 1997 National Agreement, which ensured that only trained and qualified interpreters were used, a system established following the Runciman Royal Commission due to past inaccuracies in interpreting[4]. The National Register of Public Service Interpreters (NRPSI) was created to maintain high professional standards.
5. In 2012, the Ministry of Justice (MoJ) outsourced interpreting services to Applied Language Solutions, later to Capita, despite concerns that this would lower standards. The move aimed to address administrative inefficiencies but resulted in reduced fees and dissatisfaction among interpreters. By 2024, services are managed by The Big Word (TBW) through the WordSync app, which has faced significant technical issues, leaving many interpreters unpaid for extended periods[5]. This led to industrial action and legal disputes, exemplified by the Leigh Day lawsuit against TBW for its low pay rates.
6. Additionally, 2024 has seen a rise in off-contract bookings, which offer higher rates than TBW contracts, highlighting interpreters' dissatisfaction with TBW's services and their preference for better pay and working conditions.

What are the key issues in the provision of interpreting and translation services and what impact do they have on the running of the courts, public trust, interpreters and translators.

7. The provision of interpreters in the court system faces several key challenges, including adequate training, working practices, and fees and remuneration.
8. Adequate Training: Legal interpreting requires specialised knowledge of legal terminology, courtroom procedures, and professional ethics. However, not all court interpreters receive sufficient initial training or continuous professional development (CPD), which can result in poor quality services, delays, and miscarriages of justice. Ongoing training is essential, particularly in the legal field where terminology often changes due to new legislation.
9. Working Practices: Current working practices, particularly the use of only one interpreter for spoken language during court trials, lead to fatigue, errors, and potential miscarriages of justice[6]. Unlike sign language interpreters, who work in teams to ensure quality and provide breaks, spoken language interpreters often work alone without any system for checking or correcting errors. This practice needs reform to prevent mistakes caused by mental fatigue or language nuances.
10.                     Fees and Remuneration: Court interpreters face unsustainable pay structures. Many are paid only for the time they are actively interpreting, despite spending hours on travel and waiting. This makes it difficult to earn a living wage, as interpreters can typically accept only one or two assignments per day, and overrunning court cases can further reduce their earnings. Addressing this issue is crucial to improve interpreter availability, professionalism, and overall service quality in the court system.

Is there data on the number of miscarriages of justice due to ITS error?  

11.                     Many miscarriages of justice related to interpreting errors often go unnoticed. This is because the court relies entirely on the interpreter's translation, and errors may remain hidden unless the interpreter themselves identifies and reports the issue. In some cases, defendants may notice inaccuracies but may not feel confident enough to complain. A well-known example is the case of R. v Iqbal Begum (1985), where an unqualified layperson was used to interpret, similar to the current use of "bilinguals" instead of qualified interpreters in some court hearings[7].
12.                     There are reports of trials being interrupted due to poor interpretation, with lawyers struggling to find competent replacements because of rigid contracts between the courts and interpreting agencies[8]. The courts have limited power to ensure proper service or replace inadequate interpreters, leading to delays and a lack of accountability, similar to issues seen with outsourced prison services like SERCO.

Are the required qualifications and experience of interpreting and translation services in the courts consistent?

13.                     Prior to outsourcing, the recommendation for public service interpreters was that they must be registered with the National Register of Public Service Interpreters (NRPSI), ensuring they were qualified through specific interpreting certifications like the Diploma in Public Service Interpreting (DPSI) or the Metropolitan Police Test[9]. The NRPSI maintained professional standards by requiring interpreters to follow a strict code of conduct and indicating their experience level through a grading system, allowing clients to assess risks when hiring interpreters.
14.                     However, after outsourcing was introduced, the standards for interpreter qualifications significantly declined[10]. Language Service Companies (LSCs), competing for contracts, expanded their pool of interpreters by accepting less relevant degrees, such as Language or Translation degrees, which do not adequately prepare individuals for interpreting. This oversaturated the market, reduced interpreter earnings, and diluted the profession's quality, making it less attractive for new entrants and lowering overall service standards.

Are the recommended requirements standardised across all governing bodies, contractors, and institutions?

15.                     Before outsourcing, interpreting services in the UK criminal justice system were governed by the National Agreement established in 1997[11]. This document required interpreters working in courts and police stations to be registered with the National Register of Public Service Interpreters (NRPSI) and stipulated that a registered interpreter must be used whenever possible. If no registered interpreter could be found, alternative arrangements could only be used if delays were unavoidable. Interpreters under the National Agreement were subject to a code of conduct and disciplinary procedures, ensuring high standards. Similar interpreting arrangements were in place for the UK Border Agency and immigration tribunals[12].
16.                     After outsourcing began, qualification requirements were significantly diluted, allowing Language Service Companies (LSCs) to expand their interpreter pools. This led to market oversaturation and reduced oversight, resulting in unsafe practices, including breaches of impartiality and significant errors[13]. There was no consistent regulation or accountability for the services provided, and many public service organisations lacked standardised guidance. However, some organisations continued to source interpreters directly from NRPSI, recognising the declining standards caused by outsourcing.

Are the current requirements fit for purpose?  

17.                     If by current requirements, Language Related Degrees and Diplomas, BAs in Modern Languages, BAs in Translation, Degrees in Linguistics and Degrees in Philology are meant, then the answer is NO. Public service interpreting is a skill that requires proficiency in languages as a base line. Knowledge of languages on its own does not qualify one as an interpreter because Linguistics and Philology are completely different disciplines which are not relevant to the theory of interpreting and translation.

What quality assurance and complaints procedures are in place in relation to interpreting and translation services in the courts?

18.                     We read in the Justice Committee Sixth Report of Session 2012-13: “In the course of our inquiry it was alleged that HMCTS had actively discouraged its staff from submitting formal written evidence.”[14]
19.                     There is only very limited Quality Assurance of interpreters working in the court system.  As far as it is known, it is currently carried out (but only rarely) by The Language Shop.  There are reports of lesser qualified assessors carrying out colleague assessments.

How easy is it for people to report or submit a complaint?

20.                     In the UK, complaints against court interpreters vary by source: those provided by TheBigWord are managed by The Language Shop (TLS), while off-contract interpreters fall under the National Register of Public Service Interpreters (NRPSI). NRPSI offers independent oversight and maintains professional standards, unlike TLS, which prioritises efficiency and may compromise quality and accountability.
21.                     TLS’s complaint processes, such as mystery shopping, can stress interpreters without improving service quality, leading to unethical practices like false complaints. This reliance on private companies results in inconsistent standards and fragmented complaint handling, negatively affecting access to justice for Limited English Speakers (LES) and increasing miscarriage of justice risks.
22.                     To address these issues, stronger independent regulation by NRPSI could standardise qualifications and enhance accountability, ensuring fair access to justice and restoring trust in court interpreting.

What data exists on the number and types of complaints made?

23.                     The NRPSI and The Language Shop would be a good source for this information.

How easy is it to recruit and retain skilled interpreters and translators to work in the courts?

24.                     The public service interpreting profession has become less attractive, largely due to the financial hardships caused by outsourcing, leading many qualified interpreters to leave the field. Declining recruitment statistics suggest the profession is no longer seen as rewarding, as noted by the All Party Parliamentary Group on Modern Languages (APPG)[15]. Additionally, training for interpreters may not always be suitable for legal settings, as qualifications in other fields, such as health, do not necessarily prove competence in legal terminology. One effective recruitment strategy is to source interpreters through the National Register of Public Service Interpreters (NRPSI). To retain skilled professionals, it is crucial to offer competitive and fair compensation, ensuring they remain dedicated to the profession.

What opportunities, barriers and pitfalls exist and how might these be addressed?

25.                     Before outsourcing, public service interpreting in the UK was flexible and rewarding, with interpreters feeling valued. However, outsourcing has introduced barriers such as lost direct relationships, low pay, late payments, and poor working conditions, leading many experienced interpreters to leave the profession. This has resulted in delays in court proceedings and inconsistent service quality[16].
26.                     Challenges include underqualified interpreters, administrative complexity, and a lack of oversight, fostering mistrust between clients and interpreters, which can harm vulnerable non-English speakers.
27.                     Potential solutions involve using the NRPSI to ensure qualified interpreters, improving working conditions, increasing fees, and ensuring fair treatment in complaints, which could enhance the quality and reliability of court interpreting services.

What is the potential role of new technology (such as artificial intelligence, machine translation and the digitisation of court proceedings) in the future of interpreting or translation services in the courts? (see Appendices 1,2,3)

28.                     The increasing reliance on artificial intelligence (AI) in professional settings raises concerns about a potential decline in human skill and competence, particularly in critical fields such as translation and interpreting. Despite the push for stricter accreditation and professional development, the fear is that AI may lead to the deskilling of professions and a reduction in the number of qualified practitioners. Recent technological failures, like the global IT outage, highlight the risks of fully replacing skilled professionals with AI, which could result in significant liabilities and adverse human outcomes.
29.                     AI-powered translation tools are currently deemed unreliable, especially in sensitive legal contexts such as family and immigration law, where human oversight is essential[17]. While some argue for AI's financial benefits, the costs of failure and potential errors far exceed the savings from reduced human involvement. Furthermore, machine interpreting struggles with the complexities of real-life multilingual communication, often limited to linguistic processing without understanding contextual nuances, prosody, or pragmatics. As such, AI is not yet capable of providing the quality needed for critical applications, necessitating continued human engagement and oversight in these areas[18].

Would adoption of this technology in the courts be an appropriate use? (see Appendices 1,2,3)

30.                     While AI has the potential to enhance real-time language interpretation and translation, its current limitations raise significant concerns regarding its use in legal contexts. The technology often fails to capture the nuances of legal language, cultural differences, and the importance of context, particularly in sensitive communications[19].
31.                     Key reasons against employing AI in courts include:
32.                     Rights Denied: The right to a fair trial mandates the provision of qualified interpreters for non-native speakers. Inaccurate AI translations could violate this fundamental right.
33.                     Accountability: Unlike human interpreters, AI lacks accountability for errors, which undermines the fairness and integrity of legal proceedings. Those providing the technology should be held responsible for its failures.
34.                     Confidentiality: AI and machine translation do not reliably ensure the confidentiality required in legal settings, risking violations of GDPR and privacy laws.
35.                     Personal Context: Individuals needing interpretation are often in traumatic situations, making empathetic and accurate interpretation vital. Current AI technology could exacerbate their vulnerability.
36.                     Technical Issues: AI struggles with the complexities of speech recognition, especially in less common languages or dialects. Errors in translation can have severe consequences in legal scenarios.
37.                     Cultural Considerations: Effective interpretation requires understanding cultural nuances and nonverbal communication, areas where AI falls short. Misinterpretations can lead to harm and violations of rights.
38.                     Overall, the integration of AI into legal interpretation is fraught with challenges that could compromise justice and individuals' rights.

What tools already are already in use in ITS, what form do they take and in what situations are they used?

39.                     In her article on video-mediated interpreting (VMI) in legal settings in England[20], Sabine Braun from the University of Surrey discusses challenges faced by interpreters, including inconsistent videoconferencing (VC) facilities and lack of input from interpreters in designing these solutions, which undermines their professional status. The outsourcing of court interpreter services to commercial agencies since 2012 has further worsened working conditions.
40.                     Braun notes that inadequate video links, especially in court-prison settings, are common in England, unlike in other countries where such issues are less prevalent, leading to negative perceptions of VMI among interpreters.
41.                     Although the Ministry of Justice invested over £120,000 in bespoke headsets to aid communication during COVID-19[21], this equipment remains largely unused. If compliant with ISO standards, it could enhance interpreting quality by allowing proper setups, such as tables for interpreters’ tools and working in pairs for simultaneous interpreting, reducing cognitive overload and improving effectiveness in court proceedings.

Is the current and future ITS workforce being prepared to work with technology? If so, how?

42.                     Video-mediated interpreting (VMI) is more than just a technical solution; it represents a complex social practice that is transforming the delivery of interpreting services. While VMI poses challenges, it also opens new opportunities. However, the case of England demonstrates that these opportunities can be squandered if the sociopolitical context surrounding VMI is ignored.
43.                     Academic studies emphasise two essential prerequisites for the effective use of VMI in multilingual legal proceedings. First, the involvement of the interpreting community in the design process of VMI solutions is crucial. Second, the establishment of high-quality facilities that cater to interpreters' needs is necessary for them to perform their roles confidently and accurately. Meeting these prerequisites is vital for ensuring fairness in the justice system.

What is the current capability and accuracy of market leading artificial intelligence and machine translation tools in relation to ITS? (see Appendices 1,2,3)

44.                     While leading AI technologies can process data quickly and achieve around 90% accuracy—often excelling with accents and background noise—its capabilities and accuracy should be viewed separately. AI lacks the ability to learn, adapt, and reflect on errors during interpretation, which can jeopardise court proceedings. Human interpreters excel in clarifying misunderstandings and providing the empathy and emotional intelligence necessary in sensitive environments[22].
45.                     Despite AI's utility, it cannot fully replace human interpreters in contexts requiring cultural understanding and specialised knowledge. AI should be regarded as a complementary tool, with human oversight ensuring quality. Key limitations of AI in interpreting include:
46.                     Cultural Nuances: Lacks deep cultural understanding.
47.                     Body Language: Fails to interpret nonverbal cues.
48.                     Specialised Terminology: Struggles with legal jargon, risking misinterpretations.
49.                     Ethical Concerns: Cannot adhere to ethical standards or accountability.
50.                     Confidentiality: May not securely handle sensitive information.
51.                     The best practice involves using teams of interpreters for error-checking and upholding accuracy and ethical standards.

How does this vary between languages (e.g. low resource languages or languages with relatively few written language samples), interpreting (speech to text) and translation (text to text)? (see Appendices 1,2,3)

52.                     Damian Harris-Hernandez, executive director of the Refugee Translation Project, warns against the reliance on machine translation, particularly in immigration cases, as inaccuracies can jeopardise entire cases[23]. He notes that AI-powered translation tools are especially unreliable for languages that differ significantly from English or are less documented. Language serves as a means of expressing cultural identity, and without cultural context, machine translations often reflect a Western perspective, hindering the accurate interpretation of non-English languages.
53.                     The translation process requires a trauma-informed approach, especially when working with individuals who are often traumatised; human interpreters can convey empathy and emotional nuances, which machines cannot replicate. While AI technologies hold potential benefits for society, they also raise serious ethical concerns, such as biases that can lead to discrimination and inequality, transparency in algorithm workings, and impacts on human dignity and rights. These issues underscore the need for caution and consideration in the deployment of AI in sensitive areas like translation and interpreting[24].

What capability do these tools have to deal with dialects, nuance and colloquial use of language? Request for written evidence (see Appendices 1,2,3)

54.                     Scholars in the AI field emphasise the importance of focusing on building fluency in interpreters' target languages and enhancing their strengths in areas where humans excel, such as interpersonal communication and cultural nuance. Essential skills for translators and interpreters include exceptional bilingual proficiency, analytical and critical thinking skills, a broad knowledge base, and the ability to quickly delve into new subject areas[25].
55.                     Researchers highlight that, similar to other AI tools, machine translation services often reflect and reinforce societal biases and global imbalances. For instance, the English language, with its colonial history, has abundant digital resources, while languages like Swahili and local Afghan dialects are underrepresented, making accurate machine translation challenging. This scarcity limits the ability of machines to convey messages with the same depth and cultural context as a professional interpreter.
56.                     Moreover, cultural nuances that may not translate effectively into English can lead to knowledge loss. Marek Kohn, a British science writer, asserts that while AI interpreting systems may offer cost savings, human interpreters consistently outperform machines in quality. Organisations must weigh the potential for reduced costs against the inevitable loss of interpretation quality that comes with choosing artificial intelligence over human expertise[26].

Appendix 1

Position Paper on the Use of AI in Interpreting by the International Federation of Translators

The integration of Artificial Intelligence (AI) into the field of translation and interpreting presents significant issues for both professional interpreters and the individuals whose words they interpret. As a global body representing over 65,000 individual translation, interpreting and terminology practitioners in over 65 countries, the stance of the International Federation of Translators is firmly against the replacement of human interpreters with unsupervised, unrevised machine interpreting and translation, as it creates unacceptable levels of risk and error, and undermines the role of professional interpreters and translators. 

The global trend of the previous decades has been towards increased support and investment for wider implementation of strict accreditation and quality assessment standards, increased training and ongoing professional development. We fear that increased use of artificial intelligence will severely reduce overall levels of human skill and professional competence in vital areas, leading to deskilling of professions and a marked decrease in the number of active professionals. Given the vulnerabilities evidenced during the recent technology failure due to a global IT outage, replacing skilled humans with technology entirely is a decision fraught with risk, liability, and very real consequences.

Legal Basis

The right to a fair trial is a fundamental right enshrined in law[27].  Those who do not speak the language of the court have a legal right to the assistance of an interpreter in order to be able to exercise this right in proceedings in which they participate [Article 8(2)(a) ACHR[28], 2010/64/EU[29]].

Replacing certified human interpreters with machine ‘interpreting’ via voice-to-text and text-to-voice transfer raises serious concerns regarding the accuracy, impartiality, and clarity of interpreting in sensitive legal and administrative proceedings.

Rights Denied

Given the right to a fair trial is a fundamental right enshrined in law, those who do not speak the language of the court are legally entitled to an interpreter’s assistance to exercise this right. Miscarriages of justice have occurred due to the lack of a professional interpreter[30]. Critical failure in interpreting caused by technology would similarly result in miscarriages of justice and subsequent legal challenges. MT and AI are not currently capable of replacing legal translators and interpreters.

Accountability

The use of technology in legal and administrative proceedings raises the issue of accountability. A human interpreter can be held liable for errors, but a technical device cannot. This lack of accountability for machine-generated errors poses significant risks to the fairness and integrity of proceedings. All parties to the proceedings should be liable for their actions and the implications thereof, and this includes those who supply a device being held liable for the device’s failures.

Confidentiality

Machine translation and AI cannot reliably ensure the confidentiality required in such proceedings, posing risks under GDPR and other privacy laws.

Personal Context

Individuals requiring interpreting in legal or crisis settings are likely to be highly traumatised and in a situation of high stress, lacking adequate social, family, and financial support. They are highly vulnerable, making accurate and empathetic interpretation critical. Machine versus empathetic human interpreting would compound their situation of vulnerability.

Technical Issues

Machine translation has demonstrated limited competence[31] with speech recognition complexities, such as modulation, volume, pitch, regional dialects and linguistic nuances. These errors are much greater in rare language combinations or languages with limited training data.  Even small errors can have grave consequences in legal situations. The automated transfer between languages is (and is likely to remain) far from fault-free for a considerable time in the future.[32] Machine translation is but a translation tool, not an interpreting service. It fails to adequately address cultural differences and contextual subtleties that are critical in high-stakes interpreting situations such as legal, health, emergency and crisis management and refugee settings.

Cultural Considerations

Cultural understanding and nonverbal communication are crucial in interpreting. Technology cannot adequately convey cultural nuances, body language, idiomatic expressions, and context-dependent meanings, leading to possible misinterpretations and subsequent harm and infringement of rights.

Sustainability of the profession

Experienced, professional practitioners are leaving the profession.  Issues with poor rates of pay and conditions, changes in the way individual practitioners are engaged and remunerated, and the “appification” of service provision are major contributors to job dissatisfaction and low morale.  Similarly, the use of technology as a cost-saving measure has negatively affected working conditions. If current trends continue, they “may cause a skills drain in the language industry, jeopardising the sustainability of the translator workforce and hindering the industry’s healthy growth”.[33]

Using machine interpreting as a stop-gap for a lack of interpreters is problematic. The solution to any shortage is not to fill it with untrained professionals willing to do the work for less or with technology.  While technology can assist in quality assurance in interpreting, it falls short of human capability. Effective interpretation requires more than just any form of translation; it requires accuracy and human oversight.

Conclusion

AI-powered translation is currently an unreliable technology and is not sufficiently mature to adequately replace human translators and interpreters in sensitive legal situations. Artificial Intelligence cannot be used in an unsupervised manner and requires professional human oversight both now and for the foreseeable future. Arguments in its favour that derive from financial and budgetary considerations belie a false economy, as the costs associated with failure far outweigh the cost of appropriate implementation by humans.

Appendix 2

Position Paper on the Use of AI-generated Translation and Interpreting at Court

By EULITA - European Legal Interpreters and Translators Association

On 27 September 2024

At EULITA we know that the quality of software products in our sector is increasingly better, but we also know that the accuracy level of their output is not the same as that of human translators and interpreters. We strongly believe that in sensitive areas, such as administrative, legal, medical and other settings, focus should remain on the risks involved. Even the slightest inaccuracy in translation and/or interpreting can result in serious consequences for the parties involved and we take the stance that such risk must not be taken.

Machine and AI-generated translation and interpreting are above all much faster than human translation and interpreting. However, they struggle with complex context, lack consistency, are inaccurate for lower-resource languages and not sufficiently accurate for legal, administrative, medical and other settings. The courts should be the last setting where machine and AI-generated translation/interpreting are used. If the new technologies are introduced to courts without a proper risk assessment, serious damage will be done.

Legal Basis

The right to interpretation and translation is guaranteed by the European Convention of Human Rights and Fundamental Freedoms, in particular its Article 6, which ensures the right to a fair trial by requiring that a defendant must be assisted by an interpreter if they do not understand or speak the language used at the hearing. The Directive 2010/64/EU of the European Parliament and of the Council of 20 October 2010 on the right to interpretation and translation in criminal proceedings facilitates the application of the right to a fair trial in practice. Member States shall take concrete measures to ensure that the interpretation and translation provided meets the quality required under Article 2(8) and Article 3(9).

Liability

It is worth noting that the liability issue with machine and AI-generated translation and interpreting remains unresolved. The software developers of such applications are reluctant to take any liability for their products used in the public domain, and are only willing to take liability up to the amount of the license fee for licensed applications. Our worry is that the deployers of such technologies in the administrative and legal fields, for example ministers putting forward legislative proposal[34] for technical devices to be used in the aforementioned settings, are also not willing to take on the liability for any mistakes of such tools and technical devices. Before there is a clear answer regarding the liability for both, software developers and deployers of their products, no technical device or tool should be used for interpreting or translation in administrative, legal, medical and other settings, where mistakes can lead to serious consequences.

Confidentiality

Machine translation and AI pose risks under GDPR and other privacy laws, as well, as the issue of confidentiality remains unresolved. It needs to be addressed, as in administrative, legal, medical and other similar settings personal data are always used. Therefore, it should be known in advance how they are stored and if appropriate protocols are being followed.

Accuracy

Professional legal translators and interpreters are trained in order to acquire and maintain their language proficiency in two languages (source language – SL and target language – TL), to get familiar with different fields of law and other subjects, and to learn interpreting techniques. They also abide by the code of professional ethics[35], putting emphasis on impartiality, accuracy, confidentiality, obstacles to performance quality etc. Because of this, they know where the major differences are between legal systems and can adequately translate/interpret the meaning. They are also aware of cultural differences. When interpreting, they understand the context, as well as the emotional state and the body language of the speaker, which allows them to choose the right tone of the output. Because of all that they are accurate when translating and interpreting.

It needs to be noted that the source for data of any AI and machine translation tools is the internet, where a lot of data are provided for some language pairs, in particular in combination with English, however, data which could serve as a basis for other language combinations, in particular involving the languages of lesser diffusion, are scarce or even non-existent. These tools also struggle with dialects, nuance and colloquial use of language. Sometimes they tend to convert positive statements to negative ones. They may add some phrases (hallucinations) or omit crucial parts of a sentence or an utterance. If this would happen in any part of a police investigation or court proceedings, the consequences may be devastating.

Conclusion

At EULITA we believe that even machine generated or AI generated texts/speech should be checked by a qualified translator/interpreter who is fluent in both languages (the source language and the target language) and is able to check the proposed output.

AI presents both benefits and risks. It is the role of the states to ensure that human rights, democracy and the rule of law are protected and promoted in the digital environment. We propose that any changes in the field of translation and interpreting are introduced only after thorough consultation with professional translators and interpreters and their associations.

Appendix 3

Written Evidence submitted by the FIT Europe – Regional Centre for Europe of the International Federation of Translators (FIT) on 23 September 2024

FIT Europe, the regional centre of the International Federation of Translators in Europe, represents the interests of around 40,000 professional translators and interpreters who are members of European T&I associations. This submission was prepared at the request of AIT, one of the UK-based members of FIT which falls within the FIT Europe territory.

Our internal statistics show that many such associations have individual members who are engaged in legal translation/court interpreting. We are therefore uniquely placed to respond to certain of the issues raised as part of this inquiry and to provide a European perspective on them.

FIT Europe monitors the state of the T&I sector at European level and has identified several important trends in recent years:

• a reduction in the ability of professionals to earn a sufficient income because of the rapid rise and deployment of technological “solutions” (European Language Industry Surveys/ELIS). We question whether they are in many cases actually solutions to the “problem” they seek to solve. Oftentimes the actual problem is persistent, extreme underfunding of essential state services.
• excessive hype around the capabilities of language technologies; they remain tools which require experts. The technologies themselves are still “not quite there” for important sectors such as law and medicine. There is a growing body of academic evidence confirming these points.
• the increasing adoption of a public procurement approach to essential services like court interpreting, legal translation (UK, Denmark, etc.). We note that in countries with more localised procurement the results are better though still with issues (Norway, Finland).
•                                                                                                                                                                                                                                                      poor payment practices by the companies to which such contracts are awarded (delays in payments).
•                                                                                                                                                                                                                                                      poor management of such contracts. The UK’s experience in this regard is not unique. Other European countries have faced similar issues.              
•                                                                                                                                                                                                                                                      low rates of pay for professional services, often resulting in strikes or the withholding of services (there have been strikes in the Netherlands, Austria, Poland for example).
•                                                                                                                                                                                                                                                      resistance by government authorities to increases in rates because they fail to see the value of human services having been dazzled by the promises of technological solutions.
•                                                                                                                                                                                                                                                      reduction in the level of qualifications required for court interpreting in various European countries.
• the deliberate undermining of systems of sworn translation in countries where such a system is in place by assigning translations to companies not individual sworn translators to bypass the legal rules (Poland, Austria, for example).
• insufficient numbers of professionals enrolling in the registers of courts to provide court interpreting services because of a mix of the above reasons (Greece, Italy, France, Spain).
• increasing pressures on translation/interpreting services because of migration influxes into Europe, especially since 2015, resulting in delays, inability to find interpreters, etc.
• the increasing use of free machine translation tools especially in migration settings.
• the increasing use of non-professionals (mere bilinguals, with little/no formal training) in high stakes settings (asylum interviews, court cases, etc.).
• the development of ad hoc training schemes to “upskill” and “tool up” such non-professionals to ensure persons in need of translating/interpreting are actually able to receive adequate services.
• the adoption of awareness-raising campaigns by associations about the importance of using professionals (Austria, Greece, Slovenia, Poland, for example. FIT Europe also developed materials for a European-wide campaign in partnership with various charities working in the asylum/migration sector).

In light of the above, in response to questions 6 and 7 posed by the Inquiry, the Board of FIT Europe is of the opinion that technological systems like neural machine translation (NMT) and artificial intelligence (AI) should not be deployed on their own in high stakes settings like court/police/migration interpreting or for related translations.  While there have been advancements in these technologies (especially NMT) they are still not there. Translation by AI is a mere side-effect. LLMs were not designed to translate.  Developers of such NMT/LLMs specifically recognise their limitations in this regard and include disclaimers that the output should not be trusted and that the services of a professional should be sought. Even the European Commission which has developed the e-translate tool (its own NMT system) includes such a disclaimer. MT requires caution. Raw outputs cannot be relied on is high stakes settings.

Any attempt to deploy such technologies unsupervised are to be deplored. Governments around Europe are increasingly seeking to cut costs/ “cut corners” using such tools (with such initiatives being launched in the Czech Republic, Poland and Greece at present) but in doing so they overlook concerns which ought to be central to any well-functioning legal system: ensuring that trials are fair, upholding the rule of law, providing full access to justice for all parties. Governments may be saving money in deploying such tools but this is a false economy: more is lost than gained when professionals are removed from the equation. Especially in the realm of court interpreting/legal translation the academic consensus is that such systems cannot perform comparative law operations; each legal system represents—via the medium of its own language—its own unique legal approach to the world. While such technological systems may be able to replace one word with another, that is not what a professional translator does in the legal realm; they adequately convert the meaning of the phrase into another language; they can identify areas where such conversion is difficult/not possible because of the major differences between legal systems; they can modulate the tone of the text/words to make them comprehensible to the reader/listener (for example by converting a highly legalistic text into simpler words for an uneducated before the court to grasp what is going on in the proceedings). Such systems also have a well-recognised tendency to convert positive obligations/statements to negative ones and to omit key parts of the sentence, things which are unacceptable in a legal setting. Such technologies promote the idea that a “sort of rough idea of what was said / written” is ok; in legal settings, though, accuracy and precision are key. The law operates through the medium of words; lawyers and judges choose their words with great care. Such systems unfortunately do not in their current statement of development.  System performance differs from language to language. These systems tend to perform better between major languages (say between French/German and English) and much poorer between lesser spoken languages (say between Hungarian/Maltese and English). Improvements are, of course, constantly being made, but numerous studies show that the improvements are now incremental and that a plateau of performance may have been reached. Irrespective of such improvements, the problems outlined above (omissions, etc.) have not been resolved and may not ever be resolved given the architecture of these systems. Such inconsistent levels of performance in the specific setting speak more to the need not to use such systems (or use them with great circumspection and always under the guidance of a paid expert) rather than to use them.  In response to your specific questions, nuance is frequently lost. Such tools and systems also struggle with dialects. Simply adding data does not appear to solve these problems. Using synthetic data can lead to language model collapse and worsening quality of the output rather than better quality.

Interestingly, judges all around Europe have differing opinions on the acceptability and use of such technologies in the courtroom. Our research into this matter reveals more instances of judges rejecting such technologies than accepting their use because of deep-seated (and well-founded, we would add) concerns about the reliability / accuracy of these tools. We reiterate the language of the law is the language of precision, not the language of approximation. The future of a court system reliant on such technological systems alone is a future likely to one of injustices being done. Thus, FIT Europe is of the opinion that the adoption of this technology in the courts is not an appropriate use.

Additionally, there is a growing body of evidence that the environmental implications of wide-spread use of such systems is devastating in terms of power and water requirements for the data centres and in terms of mining to procure the rare earths needed for these technologies. 

This is not to say that such tools should not be used by professionals.  One would not envisage creating scalpel-welding devices to perform cheaper, more efficient surgery on people; a qualified surgeon is always required. Similarly, a qualified professional may use MT/AI as a tool to help them with their task but the tool should not be used on its own. The risks and harms in legal settings are simply too high. Naturally since the emergence of these technologies professionals all around Europe have been engaged in training on how to use them properly. The annual ELIS surveys have consistently shown over many years now that technology is the number one topic that translators/interpreters take training courses in.

FIT Europe is of the view that no AI or MT tool is capable of ensuring the requisite decree of accuracy in relation to ITS. The cost of procuring and using such systems/tools is high; they do not solve the problem in reality. They undermine fundamentals of the legal system that the judiciary should be striving to safeguard. The government would be better served investing in human professionals who are better able to provide procedural safeguards, through better pay, recognition of status and improved working conditions.

30 September 2024